We present an analytical approach for automatic speech recognition based on the detection of acoustic cues. We have defined two levels of cue: "strong" cues both highly visible and efficiently discriminating, and "weak" cues, used in case of absence of "strong" cues. Our strategy based on the two levels of cue has the following advantages: it prevents the recognition system from proposing an inconsistent solution it permits an early decision when possible thanks to the detection of "strong" cues. We obtain nearly 50% correct early stop identification with "strong" burst cues; that validates our approach. An operational system is already running for voiceless stops followed by back vowels in continuous speech.